This study introduces task-injected layered hybrid retrieval-augmented generation (TILHR-RAG), a framework specifically designed for Persian to address the scarcity of native-language resources and the limitations of English-centric approaches. The architecture combines task-aware query augmentation, a layered retrieval strategy, and a hybrid semantic–lexical retriever, all supported by a multi-stage pipeline that includes preprocessing, document chunking, question generation, and embedding. A novel mechanism for injecting task-specific vectors directs retrieval toward domain intent while preserving comparability across documents. The layered design operates in three stages: per-task frequently asked questions (FAQ) retrieval, hybrid document search using FAISS semantic similarity combined with BM25 keyword matching, and a fallback response generated by a large language model (LLM). This structure ensures both precision and robustness. Comprehensive experiments across five progressively refined configurations demonstrate that TILHR-RAG achieves the best balance among accuracy, efficiency, and scalability, reaching 89.67% semantic accuracy with moderate latency and memory consumption on NVIDIA A100 hardware. Further evaluations on low-resource graphics processing units (GPUs) confirm that accuracy remains stable under hardware constraints, although latency increases significantly. Moreover, multilingual E5 embedding models substantially improve retrieval and generation quality for Persian – outperforming ParsBERT and Sentence-BERT (SBERT) – by mitigating challenges such as orthographic variation and complex compound word structures. Taken together, these findings establish task-injected layered hybrid retrieval-augmented generation as a practical, reproducible, and resource-efficient blueprint for Persian question answering, advancing retrieval-augmented generation for low-resource languages without requiring costly large language model fine-tuning, while also offering adaptable strategies for broader multilingual applications.

The Android operating system has a dominant market for use within a wide range of devices. Along with the widespread growth of the use of the Android system and the development of a huge number of apps for this operating system, new malicious apps are released daily by adversaries, which are difficult to identify and deal with. This is due to them using sophisticated techniques and strikes. Although there are a diverse range of classification models and risk estimation metrics for identifying malware in this operating system, there is still a requirement for more effective approaches in this context. In this paper, we present a new algorithm to calculate the security risk score of Android apps, which can be used to identify malicious apps from benign ones. This algorithm uses a novel technique named twin k-nearest neighbor. In this technique, to estimate the security risk of an unknown app, its nearest neighbors to malicious apps and its nearest neighbors to normal apps are computed separately using an appropriate distance formula. Then, the security risk of the input app can be computed using a simple formulation. In this formulation, the average distances of both k-nearest malicious apps and k-nearest non-malicious apps to the input app are used. In this way, the proposed method can calculate a high security risk for malware and a lower security risk for goodware. Experimental evaluations on real datasets show that the proposed algorithm has better performance over the previously proposed ones in terms of detection rate, precision, recall, and f1-score.

Recently Deep neural networks (DNN) have achieved a lot of success and become the most popular approach for speech recognition. DNN training for speech recognition is a difficult process due to its large number of parameters and speech dataset size. Using DNNs in a modeling task can be improved when pre-training is done using additional information. In this paper, we propose a new approach namely Gender-aware Deep Boltzmann Machine (GADBM) for pre-training of DNNs which utilizes gender information for better recognition task. The proposed pre-training method is evaluated in a phone recognition task. Experimental results on TIMIT dataset shows that the proposed method outperforms Deep Belief Network and basic Deep Boltzmann Machine.